Section: New Results
High-Order MRF models
Paticipants: Nikos Komodakis, Nikos Paragios
We developped a very general algorithm for structured prediction learning [7] that is able to efficiently handle discrete MRFs/CRFs (including both pairwise and higher-order models) so long as they can admit a decomposition into tractable subproblems. By properly combining dual decomposition with a max-margin learning method, the framework manages to reduce the training of a complex high-order MRF to the parallel training of a series of simple slave MRFs that are much easier to handle.